Re: git: 2f3f867ac6dd - main - ACPI: Add ACPI_Q_AEI_NOPULL quirk and use in EC2

From: Colin Percival <cperciva_at_freebsd.org>
Date: Sat, 26 Oct 2024 04:13:21 UTC
On 10/25/24 19:41, Jessica Clarke wrote:
> On 26 Oct 2024, at 03:35, Colin Percival <cperciva@FreeBSD.org> wrote:
>> --- a/release/tools/ec2.conf
>> +++ b/release/tools/ec2.conf
>> + # Graviton 1 through Graviton 4 have a bug in their ACPI where they
>> + # mark the PL061's pins as needing to be configured in PullUp mode
>> + # (in fact the PL061 has no pullup/pulldown resistors).
>> + echo 'debug.acpi.quirks="8"' >> ${DESTDIR}/boot/loader.conf
> 
> Should this not be restricted to arm64 in case it affects something on
> amd64?

It doesn't. :-)

> Also, is there no way to be able to dynamically set this at run
> time based on some _HID or similar? Requiring quirks in loader.conf in
> order to get a working system is less than ideal, especially if
> anyone’s trying to produce their own image, or updating an existing
> system (whether via freebsd-update, pkg or make).

This isn't the first time we've had a quirk in loader.conf (e.g. immediately
above this we have the hw.broken_txfifo quirk) but point taken.  Yes, turning
this on based on detecting Amazon's ACPI might be better -- although I'm not
entirely sure how.  The GPIO device in question has _HID=ARMH0061 which I'm
assuming applies to all pl061 devices; and probably the best we could do for
autodetection is to check a combination of "ACPI is from Amazon" and the ARM
CPU generation but I have no idea what Graviton 5 is gong to look like yet.

-- 
Colin Percival
FreeBSD Release Engineering Lead & EC2 platform maintainer
Founder, Tarsnap | www.tarsnap.com | Online backups for the truly paranoid